文心一镜微表情大模型 / ERNIE-Wizshot-ME

文心一镜微表情大模型 / ERNIE-Wizshot-ME is the second Wenxin application model Baidu formally launched with ERNIE-Light-DT-1.0 at Baidu World 2026 on 12 November 2026. National Business Daily said both models went live on Qianfan with API access for enterprise customers and partners.

Qianfan’s model list (updated 2026-11-11) has a video-generation section titled ERNIE-Wizshot-ME（文心一镜微表情大模型）. The version name is ERNIE-Wizshot-ME, the endpoint is ernie-wizshot-me, and both concurrency and queue size are 1. The docs point to “create video generation task” and “query video generation task” APIs, not the text chat API.

Model pricing (updated 2026-11-11) lists ERNIE-Wizshot-ME at 0.7 yuan for 1 second of video, with no other duration tiers in that table. Public coverage (for example unwire.pro) says the product page describes an AI video model that can make a person in an image speak naturally without a live shoot, producing 4–15 second 1080p clips at a list price of 0.7 yuan per second. Recheck the Qianfan product page and task-creation docs for duration and resolution.

Key points:

• Model list: https://cloud.baidu.com/doc/qianfan-docs/s/7m95lyy43 ; pricing: https://cloud.baidu.com/doc/qianfan/s/wmh4sv6ya ;
• Keynote coverage: https://www.nbd.com.cn/articles/2026-11-12/4645732.html ;
• This is a specific Wenxin video generation model. Do not confuse it with ERNIE-Light-DT-1.0 or with other Wenxin / MuseSteamer video models.

Why it matters

The model was named at the 12 November keynote and already appears, in both Chinese and English, on Qianfan’s official model list and pricing page, with an endpoint, concurrency cap and per-second price. That is enough for a standalone entry. It is an application video model, not a general reasoning model.

This page only restates official docs and published coverage. Claims such as 4–15 seconds and 1080p come from reports of the product page and should be re-checked upstream.

Limits

The pricing table only states “1 second / 0.7 yuan” and does not expand every duration tier. Concurrency and queue size of 1 suggest a tightly gated rollout. Do not describe it as a full replacement for live shoots, or as the same model as ERNIE-Light-DT-1.0.

1. Read the model list https://cloud.baidu.com/doc/qianfan-docs/s/7m95lyy43 and pricing https://cloud.baidu.com/doc/qianfan/s/wmh4sv6ya , and compare the keynote report https://www.nbd.com.cn/articles/2026-11-12/4645732.html ; use the latest Qianfan “create / query video generation task” docs for duration, resolution and request fields;
2. Call endpoint ernie-wizshot-me through the video task APIs, not text chat completions;
3. When quoting 0.7 yuan per second, 4–15 seconds or 1080p, name the source (pricing page vs. product page or report) and re-check upstream;
4. Do not confuse 文心一镜微表情大模型 / ERNIE-Wizshot-ME with ERNIE-Light-DT-1.0 or with MuseSteamer and other video models.